[TUHS] OSI stack (Was: Posters)

Larry McVoy lm at mcvoy.com
Thu Feb 7 09:52:57 AEST 2019

On Wed, Feb 06, 2019 at 04:40:24PM -0700, Kevin Bowling wrote:
> Seems like a case of winners write the history books.  There were
> corporate and public access networks long before TCP was set in stone
> as a dominant protocol.  Most F500 were interchanging on SNA into the
> 1990s.  And access networks like Tymnet, etc to talk to others.

Yeah, but those were all tied to some vendor.  TCP/IP was the first
wide spread networking stack that you could get from a pile of different
vendors, Sun, Dec, SGI, IBM's AIX, every kernel supported it.   System V
was late to the party, Lachman bought the rights to Convergent's STREAMs
based TCP/IP stack and had a tidy business for a while selling that stack
to lots of places.  It was an awful stack, I know because I ported it to
a super computer and then to SCO's UNIX.  When Sun switched to a System
Vr4 kernel, they bought it in some crazy bad deal and tried to use it
in Solaris.  That lead to the tcp latency benchmark in lmbench because
Oracle's clustered database ran like shit on Slowaris and I traced it
down to their distributed lock manager and then whittled that code
down to lat_tcp.c.  Sun eventually dumped that stack and went with
Mentat's code and tuned that back up.  Sun actually dumped the socket
code and went just with STREAMs interfaces but was forced to bring
back sockets.  Socket's aren't great but they won and there is no
turning back.

> TCP, coupled with the rise of UNIX and the free wheel sharing of BSD
> code, are what made the people to talk to.

BSD wasn't free until much later than TCP/IP.  TCP/IP was developed in
the 1970's.



More information about the TUHS mailing list